부산대학교 도서관

During these years, the popularity of the Internet, services such as recipe sharing, food reviews and take-out have become increasingly common in life. In the scenario of using this kind of life service, people face the problem of information overload, so people are more inclined to accept system suggestions. Recipe recommendation is pushed to the forefront. A good recipe recommendation algorithm can greatly increase the happiness index of people's lives. However, existing research in the field of recipe recommendation generally ignores the relationship between recipe attributes (such as ignoring lamb souvlaki and Korean BBQ lamb chops have the same ingredients, craftsmanship, taste, etc.) resulting in incomplete learning of user and recipe representations, which leads to a situation of poor recommendation performance. To solve this problem, this paper presents a new recipe recommendation method based on the Information Propagation mechanism of Knowledge Graph (InPKG). Specifically, this method analyzes the higher-order structural features and relational characteristics of entities through the knowledge graph higher-order propagation properties, and uses a message aggregator to aggregate the obtained higher-order characteristics to obtain a richer representation of entities. The experimental results show that InPKG has advantages over existing classical recipe recommendation models and has good interpretability.